OpenMM MCP Server

What is OpenMM MCP Server?

It is a Model Context Protocol server that integrates OpenMM molecular dynamics and Abacus DFT engines, providing a natural language interface for molecular simulations through LLM integration. It is designed for researchers, computational chemists, and developers who want to run complex simulations via conversational commands.

How to use OpenMM MCP Server?

Clone the repository, install Python dependencies (pip install -r requirements.txt), and optionally install OpenMM (conda install -c conda-forge openmm). Configure the server in your LLM client (e.g., Roo Code) by adding the provided JSON snippet to your MCP settings, pointing to run_openmm_server.py. Then use natural language commands (e.g., “Run a molecular dynamics simulation of a water molecule at 300K for 10000 steps”) to invoke tools like create_md_simulation, create_dft_calculation, or control_simulation.

Key features of OpenMM MCP Server

• Complete OpenMM integration with advanced integrators, barostats, and constraints
• DFT calculations using the Abacus engine
• Pre-configured templates for protein, membrane, and other simulations
• GPU acceleration via CUDA and OpenCL platforms
• Asynchronous task management with persistence and monitoring
• Natural language interface for complex simulation commands

Use cases of OpenMM MCP Server

• Set up a protein production simulation for 100 ns at 310 K using Amber19 and TIP3P-FB water
• Create a POPC membrane protein simulation for 50 ns with surface tension control
• Perform enhanced sampling methods like metadynamics or free energy calculations
• Run a simple water box simulation at 300 K for testing and validation
• Combine molecular dynamics with DFT calculations for hybrid QM/MM workflows

FAQ from OpenMM MCP Server

What happens if OpenMM is not installed?

The server runs in “mock mode” for testing. To run actual simulations, install OpenMM via conda install -c conda-forge openmm.

How do I integrate this server with Roo Code or Claude?

Add the provided JSON configuration to your MCP settings, replacing the path in "cwd" with the actual project directory, and include the listed tools under alwaysAllow.

Which force fields and water models are supported?

Force fields: Amber14, Amber19, CHARMM36. Water models: TIP3P, TIP3P-FB, TIP4P-Ew, SPC/E.

What compute platforms can I use?

CUDA (NVIDIA GPU, fastest), OpenCL (general GPU), CPU, and Reference (for debugging).

Which environment variables can I set?

TASK_DATA_DIR, MAX_CONCURRENT_TASKS, DEFAULT_OPENMM_PLATFORM, LOG_LEVEL, and LOG_FILE.